The Tuesday Briefing — Jul 28, 2026

The Big Picture

Here's a number worth sitting with: 53. That's how many fake software package names AI coding tools have hallucinated this week — names that don't exist, but that criminals can register and fill with malware before your developer's tool "helpfully" suggests installing one. Meanwhile, the AI coding assistants your team may already be using (Cursor, Claude Code, and others) picked up a fresh batch of serious bugs, and industry watchdogs are quietly pulling back from crowdsourced bug-hunting because AI-written spam is drowning out real findings. The pattern this week: the tools helping your team move faster are also the ones creating new, hard-to-see openings.

This Week's Top 5

1. AI Coding Tools Are Inventing Fake Software Packages — and Criminals Are Ready

What happened: Researchers found that popular AI coding assistants regularly suggest installing software packages that don't actually exist — 53 fake names across five major AI tools in one study alone. Criminals can register those exact fake names in advance and load them with malware, waiting for someone's AI assistant to recommend them.

Why it matters to your business: If your business has any custom software or an in-house developer using AI coding help, this is a quiet, easy way for malware to sneak straight into your codebase.

What to do: Ask your developer or contractor if they double-check that any package an AI tool suggests actually exists and is from a trusted source before installing it — this should be a standard step, not an afterthought.

2. Two More Serious Bugs Found in Popular AI Coding Tools (Cursor and Claude)

What happened: Security researchers found new flaws in Cursor and Claude's coding tools that could let an attacker take control of a developer's computer just by getting the AI tool to process a poisoned file — no clicking required.

Why it matters to your business: These are some of the most popular AI coding tools on the market. If a developer working on your systems uses either one, an unpatched version is a real door left open.

What to do: Ask your developer or IT contractor to confirm Cursor and Claude Code are both updated to their latest versions this week, not next month.

3. Security Bug-Bounty Programs Are Cutting Rewards Because of AI Spam

What happened: Major platforms including GitHub and Google are cutting how much they pay independent researchers who report security flaws, because so many submissions are now low-quality reports generated by AI that waste review time.

Why it matters to your business: This is a signal that the "free" security research many software vendors rely on is getting less reliable — meaning flaws may sit undiscovered or unreported longer than they used to.

What to do: Don't rely on "nobody's found a problem yet" as reassurance. Ask your IT provider whether critical business software gets any independent security testing, or just internal review.

4. A Compromised Security Scanner Shows Even Your Safety Tools Can Turn Against You

What happened: Attackers broke into and tampered with a widely used, free security-scanning tool that over 10,000 development teams rely on to catch problems in their code — turning a tool meant to protect against threats into a way to steal cloud passwords instead.

Why it matters to your business: If your developer or IT provider uses free or open-source security tools (many do), this is a reminder that those tools need their own updates and checks too.

What to do: Ask your developer which security scanning tools they use, and whether they're subscribed to get alerted if one of those tools is ever compromised.

5. Ransomware Gang Hit 7 Small Businesses in a Single Day

What happened: One ransomware group posted seven new small-business victims in a single day this week — nearly 40% of all new ransomware victims reported that day — continuing a trend of criminals specifically hunting for smaller companies with weaker defenses.

Why it matters to your business: Attackers are increasingly choosing targets based on how exposed and under-defended they are, not how much money they have — meaning "we're too small to be a target" is no longer true, if it ever was.

What to do: Confirm your business has three basics in place: multi-factor authentication (a second login step) on all important accounts, working backups that you've actually tested, and a way to detect unusual activity — not just antivirus software.

Quick Hits

• Microsoft's July software update fixed a record 1,449 separate security flaws in one batch — a reminder of just how much needs patching behind the scenes each month.

• A widely used Chinese software library (Fastjson) has a serious flaw being actively exploited right now — ask your developer if any of your systems use it.

• Regulators in Europe pushed back the toughest AI rules to 2027 and 2028, but added new bans effective this December — worth knowing if your business operates internationally.

• Researchers found over 7,600 fake code repositories on GitHub, with more than 800 disguised as legitimate AI add-ons — another reason to only install add-ons from verified sources.

• A survey found 62% of small businesses faced an AI-powered attack last year, and 82% of phishing emails now use AI-written text — a good excuse to remind your team to slow down before clicking.

• Anthropic released a new flagship AI model (Claude Opus 5) specifically tuned to be harder to misuse for hacking — a small sign vendors are starting to build in safety by default.

• A batch of newly disclosed weaknesses (called "the authorization gap") shows many popular AI-agent tools don't properly check permissions before letting the AI take actions like sending files or running commands.

One Thing to Do This Week

Ask your developer or IT provider this exact question: "When our AI tools suggest installing new software or code libraries, does anyone verify those suggestions before they get used?" This week's research found AI tools regularly recommend software that doesn't exist — and criminals are pre-loading those exact names with malware. It's a five-minute conversation that closes a door most businesses don't know is open.

Worth Reading

• Slopsquatting Explained — How AI tools inventing fake software names is becoming a real supply-chain risk.

• The Trivy Scanner Compromise — A plain-language look at how a trusted security tool got turned into an attack vector.

• GitHub's Bug Bounty Overhaul — Why major platforms are paying less for security tips, and what it means for software safety overall.

• Ransomware Victim Tracker — A running list of which industries and company sizes are getting hit hardest right now.